Candidate Screening Tools: Compare Decisions, Not Just Features

Candidate screening tools help employers decide which applicants move forward. They can apply eligibility rules, compare resumes with job requirements, collect screening answers or score assessments. Not all use AI, and not all make decisions automatically.

The most useful buying question is: What evidence does this tool use, and what happens to a candidate because of its output? A summary for a recruiter, a ranked review queue and an automatic rejection need different controls.

This guide focuses on purchasing checks, not vendor rankings or endorsements. Product examples and selected U.S. rules references are checked as of October 30, 2026. For the underlying mechanisms and application advice, see how AI candidate screening works.

Choose the screening function first

Start with the bottleneck in your current process—not a request for “AI screening.” Hiring technologies can evaluate qualifications, score resumes, conduct interviews and administer skills tests, as the U.S. Department of Justice explains.

Screening function	Input and output	What to check before buying
Eligibility or knockout rules	Application answers trigger a pass, flag or rejection	Are the requirements genuinely essential? Can errors be corrected?
Resume matching	Resume evidence is compared with role criteria to produce scores or rankings	Can reviewers inspect the evidence, weights and missing information?
Screening chat or interview	Candidate answers become transcripts, summaries or evaluations	Is the tool collecting facts, scoring answers or both? What happens if the session fails?
Skills assessment	A test or work sample produces results against a scoring rubric	Does it measure relevant job skills, and can candidates access it?

An applicant tracking system may contain several of these functions, or connect to separate products. Compare the actual configuration you would deploy. For a deeper comparison of tests and work samples, see candidate assessment tools.

Separate scoring from automatic rejection

Two documented Greenhouse functions show why the distinction matters:

• Auto-reject lets an employer attach a rejection rule to a particular application-question response. Greenhouse’s example is a required Class A commercial driver’s license. The rule can assign a rejection reason and send an email. It does not depend on an AI judgment of the resume. Greenhouse’s documentation also says new-application notifications are not sent for auto-rejected applicants.
• Talent Matching compares resumes with a recruiter’s defined, weighted criteria and provides match scores, supporting highlights and a breakdown. Greenhouse says this feature does not automatically reject or advance candidates; unreadable resumes and opted-out applications are marked for manual review. These are documented product behaviors, not independent evidence of hiring accuracy. Source: Greenhouse.

Ask vendors to demonstrate the whole path from application to disposition. A feature that only ranks applicants can still limit consideration if reviewers never reach the bottom of the queue. “A human makes the final decision” is not enough: specify who reviews low-scoring and unscored applications, and when.

Six checks for a screening-tool shortlist

1. Approve the criteria before scoring applicants

Write down essential requirements, preferred qualifications and acceptable equivalent experience. Ask whether the product distinguishes “not found in the application” from “candidate does not have it.” Inspect AI-generated criteria before enabling them.

Request validation evidence relevant to the intended position and purpose—not just a general accuracy claim. The EEOC’s selection-procedure guidance recommends appropriate validation and stresses that vendor documentation does not remove the employer’s responsibility.

2. Trace a result back to its evidence

In a demo, ask the vendor to show the original answer or resume passage behind a score. Use fictional applications to test an ambiguous qualification, a parsing failure and a corrected application.

Require a clear route for correction or manual review. A polished explanation is not useful if reviewers cannot check it against the underlying material.

3. Test decision controls and records

Confirm who can change criteria, weights and thresholds; whether changes affect existing applicants; and how decisions are recorded. Ask to export the criteria, relevant inputs, outputs, timestamps and reviewer actions.

Also test what happens when the integration fails. Require a visible exception and recovery route so that an assessment result arriving late does not silently become a failed assessment.

4. Make accessibility part of the workflow

Test screen-reader access, keyboard navigation, time limits and any audio or video requirements. Identify the person responsible for accommodation requests and explain that route before candidates begin.

For covered U.S. employers, the ADA requires reasonable accommodations for qualified applicants with disabilities during hiring unless they would create undue hardship. The DOJ guidance also explains that using another company’s technology does not excuse disability discrimination.

5. Check data handling and total cost

Ask what the vendor retains: resumes, answers, recordings, transcripts, scores and logs. Confirm deletion behavior, subprocessors, training reuse and any differences between deleting data in the ATS and deleting vendor-held copies.

Compare costs using expected applicant volume, including incomplete sessions, reassessments, integrations, implementation and reviewer time. A low per-screen price is not the whole operating cost.

6. Verify obligations for the actual use

Do not treat a vendor’s “compliant” label as a legal determination. Requirements depend on jurisdiction, tool design and how the employer uses it.

For covered AEDT use in New York City, Local Law 144 requires a bias audit within the preceding year, publicly available audit information and required notices. Enforcement began July 5, 2023. Source: NYC DCWP. DCWP also says the law prescribes no specific action based on audit results; separate anti-discrimination laws still apply. An audit is therefore not a guarantee of lawful selection. Source: DCWP FAQ.

These are selected U.S. requirements, not a comprehensive legal checklist. This is informational, not legal, HR or employment advice; confirm applicable requirements with qualified counsel.

Pilot before letting the tool exclude candidates

For an initial pilot, keep the tool advisory. Have reviewers assess applications against the same approved criteria, preferably before seeing tool scores. Review disagreements—including qualified applicants the tool would have filtered out—not only the top-ranked results. Human ratings are a comparison, not unquestionable ground truth; investigate the evidence behind both judgments.

Track reviewer time, missing or incorrect evidence, manual-review volume, completion failures and selection-stage disparities where lawful and appropriate. Set pause conditions before launch, then assign an owner for monitoring after deployment. A small pilot can expose workflow problems; it cannot establish universal accuracy or absence of discrimination.

For candidates: answer eligibility questions accurately, check any application preview for parsing errors, and ask how to correct missing information. Before a chat, interview or test, ask what is evaluated, how technical failures are handled and how to request an accommodation. Do not assume an immediate rejection proves that AI scored your resume.